Specifically, pod.Start() always returned CtrErrors, even if none failed. This cause podman start to not return the successfully started pod id.
Also, pod.Kill() didn't return an error along with ctrErrors.
Signed-off-by: haircommander <pehunt@redhat.com>
Closes: #1272
Approved by: rhatdan
Also, don't return the internal podState struct - instead return
a public inspect struct.
Signed-off-by: Matthew Heon <matthew.heon@gmail.com>
Closes: #1258
Approved by: rhatdan
Currently we add mounts from images, volumes and internal.
We can accidently over mount an existing mount. This patch sorts the mounts
to make sure a parent directory is always mounted before its content.
Had to change the default propagation on image volume mounts from shared
to private to stop mount points from leaking out of the container.
Also switched from using some docker/docker/pkg to container/storage/pkg
to remove some dependencies on Docker.
Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>
Closes: #1243
Approved by: mheon
This removes anything but structs and simple accessors from
pod.go itself, which is a target file for FFJSON generation. This
should reduce the amount of times FFJSON needs to run.
Signed-off-by: Matthew Heon <matthew.heon@gmail.com>
Closes: #1247
Approved by: rhatdan
During refresh, we cannot hard-fail, as that would mean leaving a
partially-configured state behind, leaving libpod unable to start
without manual intervention.
Instead, log errors refreshing individual containers and pods and
continue. Individual containers and pods may be unusable and need
to be removed manually, but libpod itself will continue to
function.
Signed-off-by: Matthew Heon <matthew.heon@gmail.com>
Closes: #1252
Approved by: rhatdan
When checking if the container has already been removed, use
c.state.HasContainer() instead of the runtime's API to avoid
trying to take the already acquired lock.
Fixes: #1245
Signed-off-by: Valentin Rothberg <vrothberg@suse.com>
Closes: #1251
Approved by: baude
If the stdout and stderr are not attach, podman will at least get
a messsage that the container has completed and finish.
This fixes the
`podman run -a stdin fedora true`
Hang issue.
Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>
Closes: #1239
Approved by: mheon
...and some naming decisions.
This change ensures podman save doesn't incorrectly prepend localhost when saving an image.
Signed-off-by: haircommander <pehunt@redhat.com>
Closes: #1140
Approved by: rhatdan
When pulling a fully-qualified image that fails, we should not be talking about
registries/search registries in the the error message as it is not applicable. If
a image that is fq'd and fails to pull, the error should be simplified.
```
$ sudo podman pull this-does-not-exist.example.com/foo
Trying to pull this-does-not-exist.example.com/foo...Failed
error pulling image "this-does-not-exist.example.com/foo": unable to pull this-does-not-exist.example.com/foo: unable to pull image, or you do not have pull access
$
```
Resolves: #1212
Signed-off-by: baude <bbaude@redhat.com>
Closes: #1216
Approved by: mheon
It will be handy to know the runc and conmon versions as our
code gets into the wild.
Signed-off-by: baude <bbaude@redhat.com>
Closes: #1207
Approved by: rhatdan
when pulling, we can fail to find an image (i.e. it doesn't exist) or we
can not have authority/access to pull it. the registries don't tell us
one way or another so the error message needs to cover both.
Resolves#1194
Signed-off-by: baude <bbaude@redhat.com>
Closes: #1195
Approved by: rhatdan
Again, we only needed them split for tests; so, integrate them back.
Then drop all remaining references to pullRefName and pullGoalNames,
which are not used for anything.
Should not change behavior
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1198
Approved by: mheon
This more or less reverts 9c9401a96c
"Replace getPullRefPair with getPullRefName"; now that tests don't require
us to use pullRefName, move creation of storage references deeper into the
calls stack to reduce string use.
ir.getSinglePullRefNameGoal is accordingly updated to ir.getSinglePullRefPairGoal,
and we need to add a ~duplicate singlePullRefPairGoal; that duplication
of singlePullRefNameGoal will soon be resolved by dropping singlePullRefNameGoal.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1198
Approved by: mheon
Now that we don't need a separate pullGoalNamesFromImageReference for
running tests, inline it back.
This forces us to add some glue code to getSinglePullRefNameGoal
and to convert between pullGoal and *pullGoal; that is temporary
and will be cleaned up soon.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1198
Approved by: mheon
This merely wraps the
> return singlePullRefNameGoal(getPullRefName(... reference)), nil
pattern which is used for almost all getPullRefName uses. For now
it seems not really worth it, but it will result in shorter code
(and smaller migration) after we replace getPullRefName with
getPullRefPair, which can fail, again - the pullGoalNamesFromImageReference
will not have to add any error handling.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1198
Approved by: mheon
Similarly to pullGoalNamesFromImageReference, use a storage.Store and
test the actually created references; that is more representative,
and clearly shows the impact of further normalization in
storageReference (like defaulting to :latest on NameOnly references).
Only modifies tests, so does not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1198
Approved by: mheon
pullGoalNamesFromImageReference has been added only to allow testing without
a storage.Store, because I thought that a storage.Store can only be created
by root.
It turns out that non-root stores, at least good enough for reference
parsing and formatting, are possible (and have existed in c/image/storage
tests), so this creates such a store, and modifies the existing
test to test the created c/image/storage.storageReference values
instead of strings; that is more representative, and clearly shows
the impact of further normalization in storageReference (like
defaulting to :latest on NameOnly references).
Eventually we will want to get rid of pullGoalNames / pullRefName.
Only modifies tests, so does not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1198
Approved by: mheon
This is similar to the PushImageToHeuristicDestination RFC.
The goal is to be very explicit about which functions try to heuristically
guess what is the expected format of the string. Not quite "shaming"
the users, but making sure they stand out.
RFC:
- Is this at all acceptable? Desirable?
Should not change behavior (but does not add unit tests).
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
FINALLY, (podman load) can pass through an ImageReference directly from
loadCmd all the way to pullGoalNamesFromImageReference, making sure not
to trigger the docker-like reference parsing heuristics.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
They are not used anywhere in the packagee.
Two of the values still have users in the CLI, but used only once.
So, use the .Transport.Name() calls in there directly, that is
likely to be cheaper (and makes the files depend directly
on the transports instead of referring to them indirectly through
libpod).
RFC: Should not change behavior in _this_ repo, but it is an
externally-observable API change. Is there any user that could
notice?
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
All callers of LoadFromArchive expect the input to be in the
transport:name format, or create it that way. So, pass a
types.ImageReference instead of a string.
That requires us to add an explicit parse step in (podman pull);
in (podman load) we can, instead of pasting strings, create
native objects directly.
Changes the error behavior of (podman pull), we no longer
try heuristically parsing docker-archive:... inputs as
Docker references.
Also changes the string reported by (podman load) if all parsing
attempts fail, to be only the path instead of dir:path. The error
message itself is likely to be the same (from directory.Transport).
(While at it, update a mismatched comment.)
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
Now that we have a pullGoal, separate determination of the goal from
performing it; we will then introduce another entry point with
a supplied types.ImageReference.
Also remove or correct some misleading comments.
Should not change behavior (but does not add unit tests).
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
Use the parent types.SystemContext data instead.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
Use the parent types.SystemContext data instead.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
Use the parent types.SystemContext data instead.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
All callers of getCopyOptions also call GetSystemContext with the same three parameters;
we will want to simplify this by passing the first SystemContext to getCopyOptions,
which can then inherit this data instead of so many parameters everywhere.
For now, just add a *types.SystemContext parameter without using it.
Should not change behavior (but does not add unit tests).
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
pullImage (now) only uses Image.InputName; it is really used to _create_
an Image object, based on the pull results (as is most visible in the
LoadFromArchive caller), so it should not be a method on it.
This also simplifies a bit the number of different kids of uses of
Image.InputName; still apparently not enough to clearly document
the field, though.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
... when we even only count them.
This eliminates a rare error case, and saves time re-reading and re-parsing
the input.
(We still compute registryPath redundantly, and it may get out of sync.)
Should not change behavior (but does not add unit tests).
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
Instead of duplicating the hasRegistry logic, just record whether we
did use search or not.
Should not change behavior (but does not add unit tests for all of it).
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
Instead, encode it explicitly in pullGoal.pullAllPairs.
Should not change behavior (but does not add unit tests for
all of it).
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
This is an intermediate version of pullGoal, which exists basically
only for easier testing without containers-storage: (i.e. root access)
in unit tests.
Like pullGoal, we will add more members to make it useful in the future.
RFC: Unlike pullGoal, the return value is *pullGoalNames, because there are
quite a few (return nil, err) cases which would be more difficult to read
when returning a value.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
The eventual goal is to cleanly capture semantics like "pull all images
for DockerArchive" and "did a search through $registries" without
hard-coding it through; and to allow a pullImage variant where
the caller can pass an imageReference directly.
For now, this just wraps []pullRefPair and should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
We are passing the values, don't really need the pointer sharing semantics,
and the structures are small enough, and the arrays short enough,
that we very likely lose on the indirect accesses more than we save on
quicker copying of the slices when extending them. Value semantics
is safer anyway.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
We are passing the values, don't really need the pointer sharing semantics,
and the structures are small enough, and the arrays short enough,
that we very likely lose on the indirect accesses more than we save on
quicker copying of the slices when extending them. Value semantics
is safer anyway.
Should not change behavior.
Signed-off-by: Miloslav Trmač <mitr@redhat.com>
Closes: #1176
Approved by: rhatdan
Matthew Heon
haircommander
Daniel J Walsh
Valentin Rothberg
baude
Miloslav Trmač